5ERASI: 5G Service-Enabled Resource Allocation and Sub-Slice Isolation for Open RAN

Network slicing supports diverse applications from commercial services to Mission-Critical Services (MCS) such as disaster management. While the Open Radio Access Network (O-RAN) extends slicing through intelligent control in the Radio Access Network (RAN), current 3GPP specifications only define resource allocation at the slice level, without procedures for sub-slice management. This limitation poses a barrier in disaster-oriented MCS systems, where heterogeneous sensors, applications, and services coexist within the same slice. Our approach addresses these challenges by introducing 5ERASI (5G Service-Enabled Resource Allocation and Sub-slice Isolation), a Machine Learning-driven (ML) sub-slice orchestrator that enables per-service resource allocation, isolation, and dynamic policy enforcement. Our evaluation suggests that 5ERASI incurs a minor overhead, increasing User Equipment (UE) initial access time by an average of 23.41 ms, even in multiple-UE scenarios. Comprising Principal Component Analysis (PCA), our ML classification achieves 98.42% accuracy, 19.40 ms prediction time, and 631.58 ms training time, making it a lightweight and effective algorithm for offline classification while keeping 97.96% accuracy in an online setting. In the online slicing evaluation, the isolation level can be maintained at 2.03%, which is 0.85% below the baseline threshold of 2.88%. Finally, leveraging the RAN Intelligent Controller (RIC) architecture, 5ERASI enables dynamic, on-the-fly RAN resource allocation while preserving traffic isolation, and empirically demonstrates that the proposed additional signaling procedure can be feasibly integrated into the 3GPP framework without disrupting the existing 3GPP Initial Access Procedure.